Research on Multi-class Fine Motor Intention Recognition of Hands

To address the problem of poor precision of fine motor intention recognition, which is difficult to apply to the field of neural rehabilitation, this paper designed a bilateral motor imagery (MI) paradigm using dynamic visual imagery: There were four kinds of motor intentions: left hand to finger, left hand to grip, right hand to finger, right hand to grip. Time-frequency map, brain topographic map and R2 coefficient were used to analyze the divisibility features among different categories. FBCSP (filter bank - common space model) and SVM (support vector machine) models were used to classify the fine motor intentions among various categories. The research results prove that the event-related desynchronization (ERD) features of multiple types of fine motor intentions in the same part are separable under the dynamic visual auxiliary stimulus paradigm. The average recognition accuracy of 4 types of fine motor intentions of 12 subjects can reach 63.7±3.67%, which is significantly higher than the random probability of 4 types. The experimental results show that the proposed method can effectively distinguish multiple types of fine motor intentions in the same part, and provide a new way for its application in the field of neurological rehabilitation.